José Valim

The problem is that the error message ends up being generic: "expected tensors to have matching leading dimensions" while today we can say "expected the data to have the same...

Btw, @krstopro, would you like to join the Scholar team alongside @msluszniak? No additional commitment is required from your side, it is mostly to recognize your contributions so far. :)...

Invited. Welcome @krstopro!

There is no priority per-se, whatever you feel is the most useful to you (or you are personally interested on).

Here is a small patch for the first one, but unfortunately it is not enough, so perhaps something else is wrong: ``` diff --git a/lib/scholar/cluster/hierarchical.ex b/lib/scholar/cluster/hierarchical.ex index 481b6cd..562367f 100644 ---...

I have commented Ward for now, see 6845727ee7889d085a9d79cec948dcf3c94ed2bc.

There is another option here to remove the duplication: 1. Remove the fit_transform functions from `Scholar.Preprocesing.*` modules 2. Keep the current functions in `Scholar.Preprocessing` as the `fit_transform` variant But I...

Wouldn’t it make more sense to have a build of MLIR without XLA? :)

If they export all symbols, I guess they would conflict with each other when loading the .so? I am asking because we might be going on a similar path soon....

I am not a big fan of this approach. The current implementation is too complex by trying to support three different formats and it does not work on Windows. Perhaps...